Moderation effect on non significant relationship

[JWB]

Hi everyone,

I'm having a model with with the IV (A) influencing the DV (B). The path is non significant. Then I add the moderator C to moderate the relationship between A and B. The path stays non significant, however, the moderating effect is significant.

How do I interpret this result? Is it even possible to interpret a moderating effect on a non-significant path?

Thank you very much.

Kind regards
JWB

[jmbecker]

Sure it is possible to interpret a moderating effect on a non-significant path. If your interaction term is significant then you can conclude a moderation effect.
The path from A to B is non-significant at the mean of the moderator. Changing the level of the moderator to + or - 1 standard deviation or even less may make the conditional effect significant.

It is always good to rewrite your equations to illustrate interaction effects.
B = a +beta1 A + beta2 C + beta3 AxC
<=> B = a + beta2 C + (beta1 + beta3 C)A

The effect of A on B is beta1 (the direct effect) + beta3 C (the Interactioneffect). Hence, the effect of A on B varies with different levels of C (That is the nature of moderation.)
If C is 0 (in case of mean centering or standardization at the mean of C) then A has only its direct impact beta1. However, if C is larger or smaller than 0 (larger or smaller than its mean) than you add beta3 times the value of C. For example, for one standard deviation above the mean you add beta3. beta1+beta3 may be significant or even beta1+0.5*beta3. You can also perform signficance test for these conditional effects, but you have to calculate them on your own using Excel and the Samples output of the Bootstrapping results for the path coefficients in SmartPLS.

[Yusarina]

Hi. I am having the same issue.
X - Y - not sig (p-value = 0.322) (beta = 0.068)
M- Y - not sig (p-value = 0.460) (beta = 0.105)
But when I include moderating effect, it gives p-value = 0.078 (beta = 0.109)

I am concluding that there is moderation effect, sig at 10%. My understanding - M on its own is not significant to affect Y, but when X and M interact, together their impact is sig on Y. Is my understanding correct?

My conclusion is challenged based on the fact that how could there be moderation effect if both X-Y and M-Y are not sig.

Your view is appreciated. Tq.

[jmbecker]

I am concluding that there is moderation effect, sig at 10%. My understanding - M on its own is not significant to affect Y, but when X and M interact, together their impact is sig on Y. Is my understanding correct?

Not quite correct. You have a significant moderating effect. That implies that the strength of both X and M vary when the other varies. If you have high levels of M your have a stronger effect of X, which might be significant. If you have high levels of X you have a stronger effect of M, which might be significant. Whether they are significant depends on how much you increase M or X above the mean and whether that values is still a value commonly occuring in your data.
To know whether they are significant you need to use your bootstrapping results and calculate the conditional effects at varying levels of the moderator (usually + or - one standard deviation from the mean).

[Yusarina]

Thank you Dr Becker for your reply.

So, the direct relationship between X - Y does not necessarily significant for moderating effect to be significant, right?

To proceed, I have to do as your suggestion - "To know whether they are significant you need to use your bootstrapping results and calculate the conditional effects at varying levels of the moderator (usually + or - one standard deviation from the mean)".

Would you mind to suggest any articles that can guide me on how to proceed to determine the moderating effect?

Your help is much appreciated.
Thank you.

[awwad]

in this case the moderating effect is "full mediation" on the other hand if there is a direct effect between IV and DV then the mediation effect is "partial mediation". see Hair et al. 2003.